Hierarchical method to classify emotions in speech signals

Abstract

Recently studies have been performed on spectral features such as Mel Frequency Cepstral Coefficients (MFCC) and Linear Predictor Cepstral Coefficients (LPCC) for speech emotion recognition. It was found in our study that the Fourier Transform of MFCC time trajectories also play an important role in speech emotion recognition. And also a new hierarchical classification method was proposed based on K Nearest Neighbour (KNN) and Fisher Linear Discriminant Analysis (FLDA). This reduces the computational cost drastically because a lesser number of feature set is used. Also this methodology helps in resolving the non-linearity of the clustering problem. The emotions anger, sadness, disgust, neutrality, boredom, fear and happiness were recognized with the accuracies of 95%, 100%, 85%, 86%, 77%, 53% and 80% on the standard German database(EMODB).